CN104574271B - A kind of method of advertising logo insertion digital picture - Google Patents

Abstract

The invention belongs to the technical field of image editing, in particular to a method for embedding an advertisement icon into a digital image. Image fusion is an important branch in the field of computer vision. It achieves smooth transition of images by determining the value of each pixel in the overlapping area of two images to be fused. The method of the invention not only considers the value of the pixels in the overlapping area, but also focuses on the analysis of the image content, and achieves the best effect of image fusion by selecting a suitable fusion area and detecting the relevant information of the area. Experimental results show that this method achieves the insertion of advertisement icons well, making the advertisement icons appear natural and eye-catching on the background image.

Description

A method of embedding an advertising icon into a digital image

technical field

The invention belongs to the technical field of image editing and relates to an image fusion method, in particular to a method for embedding an advertisement icon into a digital image.

Background technique

Traditional image fusion technology is basically studied as an important branch of multi-sensor information fusion. It is a modern high-tech that integrates multiple disciplines such as sensors, image processing, signal processing, computers and artificial intelligence. The main idea of this image fusion is to use a certain algorithm to synthesize multiple source images from multiple sensors into a new image, so that the fused image has higher credibility and less uncertainty. and better understandability. Image fusion is not simple superposition, it produces new images with more valuable information. At present, this type of image fusion technology has shown great application potential in the fields of automatic target recognition, computer vision, remote sensing, robotics, medical image processing, and military applications.

In the field of image fusion involved in the method of the present invention, a new image is generated by embedding an object or a region in the source image into the target image. For the new composite image to look realistic, the borders should be seamless and natural. This kind of image fusion technique is a basic problem in computer image processing.

Image fusion can be achieved by weighted average algorithm, multi-resolution method or fusion method based on gradient domain. Because the weighted average algorithm only performs weighted average on the overlapping area, it only realizes the fusion transition in the overlapping area, it is difficult to eliminate the synthetic ghost, and it is very sensitive to the registration error. However, the multi-resolution stitching method decomposes the image into multiple scale images and then synthesizes them, which can not only realize the fusion transition within the whole image range, but also reduce the sensitivity to registration errors. Its shortcoming is that multiple filtering will This causes the signal to weaken, so the final composite image will be dark and blurry. The fusion method based on the gradient domain is to use the gradient field to realize the synthesis. Since the gradient field of the overlapping area needs to be calculated, the synthesized image will not appear darkening and blurring.

Prez et al. proposed the Poisson image editing method ^[2] in 2003. This method uses the image gradient field to guide the interpolation of the area to be fused, and reduces the image fusion problem to the minimization of the objective function, and uses the Poisson equation to solve this problem. Variation problem.

Poisson image editing solves a Poisson equation according to the boundary conditions specified by the user, and realizes continuity on the gradient domain to achieve seamless fusion of colors at the boundary. But if the two regions have different texture detail features, the boundary of the edited region is still obvious. Burt et al. improved Poisson image editing and proposed a multi-resolution technology based on splines to achieve seamless fusion of two images by interpolating images at different levels ^[3] . An optimization method proposed by Jia et al. can automatically find the optimal boundary and simplify the interaction ^[4] . This method divides the object of interest from the area of user interaction through GraphCut, and finds the optimal boundary between the boundary of user interaction and the object of interest.

Contents of the invention

In order to overcome the deficiencies in the prior art, the purpose of the present invention is to provide a method for embedding advertising icons into digital images, which is mainly aimed at the fusion of advertising trademarks (trademarks, etc.) and background images, which enhances the final image by automatically searching for the best insertion area Blend effects and simplify interactions.

The invention provides a method for embedding an advertisement icon into a digital image, the specific steps are as follows:

(1) The best area selection of the background image

Embed the advertising icon into the digital image, the advertising icon is the insertion object, and the digital image is the background image; select the local area in the background image that has no strong edges, low energy, and a large color difference between the color and the insertion object as the candidate area; and then insert The color information of the object and digital image, the energy information and edge information of the background image, select the best area in the background image; its best area Satisfy the following formula:

Among them: V _i represents the suitability of the i-th area in the background image, and the larger the value, the better the effect of area i as an insertion area; E _i refers to the energy information of the background image of the i-th area, and the energy information is more in line with the selection criteria , the larger the value; B _i , C _i represent edge information and color contrast information respectively; α, β, γ are artificially selected weight parameters, and it is recommended to set them to 0.33.

(2) Affine transformation of the inserted object

First use edge detection to generate the edge image of the background image, and then use the Hough transform method to extract the straight lines in the background image, select the most relevant straight line in the straight line detection map of the background image, and calculate the parameters of the affine transformation to The direction of the straight line is the standard for the affine transformation of the inserted object, that is, to transform the inserted object in terms of angle, so that the inserted object fits the best area of the background image;

(3) Image Fusion

The image fusion method based on Poisson's equation is used to fuse the inserted object obtained after affine transformation with the background image at a suitable position. The method of Poisson fusion can be found in reference [1].

In the present invention, the local area without a strong edge is searched from the edge image, and the threshold value is determined to find a relatively smooth area to prevent the inserted object from crossing multiple objects in the background image; the local area with lower energy is obtained from the energy map The threshold is set to prevent the insertion object from being placed on an important object, and there will be obvious traces of manual processing; the local area with a large color difference is directly obtained by comparing the color information of the background image and the insertion object, in order to make the insertion Objects should be as conspicuous as possible.

In the step (1) of the present invention, select a local area in the background image that has no strong edges, low energy, and a large difference between the color and the color of the inserted object as the candidate area; specifically, it is to make the inserted trademark look more natural ,

Objects inserted first cannot span multiple entities, or severe distortion will result. This method preliminarily determines whether there are multiple objects in the area by detecting the edge of the object in the background image. If there is strong edge information in the selected area, it is very likely that the object will cross the entity after insertion, and such an area should be excluded. .

Also, inserting an object into an important area of the background image can easily distort the new image because the inserted logo will affect the theme of the background image. For example, in some images, the entity in the important area is the human face, and the trademark attached to the human face is obviously too unreal, and it is difficult to choose the angle and shape to fit. This method uses the energy map to select relatively minor areas in the background image for interpolation.

Finally, this method also takes into account the color differences of the inset object and the inset region. If the colors of the two are similar, after using the Poisson method to fuse, the inserted object is easy to blend with the background. This is the goal of Poisson fusion in essence, but the insertion of the advertising icon (such as a trademark) targeted at this article makes the advertising icon eye-catching The actual application goals do not match. Therefore, this method selects regions with large color differences as candidate regions.

In step (2) of the present invention, the most relevant straight line should be located within a certain range from the selected region RE, which is three times the distance between the selected region RE and the nearest line segment. When selecting the most relevant straight line, comprehensively consider the length of the straight line and the distance from the best background region RE, the closer the straight line is to the selected area, the greater the correlation; the longer the straight line, the more representative, the more relevant. Sex is also greater. After having the most relevant straight line, the affine transformation can be performed on the inserted object according to the direction of the line segment.

In the present invention, before the advertisement icon is embedded into the digital image, the advertisement is subjected to icon preprocessing; the preprocessing steps are as follows:

(1) Process the edges of the advertising icon so that the advertising trademark does not contain irrelevant redundant pixels;

(2) Scale the advertising trademark. After scaling, the size of the inserted object should not exceed a quarter of the size of the background image.

The beneficial effect of the present invention is that: the method proposed by the present invention not only considers the effect brought by the fusion method during image fusion, but also considers the influence of the content of the background image on the fusion effect, and is a content-based image fusion method. The present invention is mainly aimed at the insertion of advertising icons, such as trademarks, and can be applied to advertising. This method automatically searches for a suitable insertion area in the background image, and transforms the insertion object, enhances the final fusion effect, and simplifies the interaction.

Description of drawings

Fig. 1 is a flowchart of the present invention.

Figure 2 is the result of using this method to insert the Ford logo into the background image.

detailed description

For an image to be inserted and a background image, the method described in Figure 1 can be used for implementation.

The specific implementation method is:

(1) Firstly, preprocess the inserted image. Taking the trademark as an example, try to make the area of the inserted image except the trademark itself white, so as to clarify the actual inserted object and avoid inserting redundant information. If not, the insertion object in the image is manually selected so that it contains as few redundant pixels as possible. Then scale it to a suitable size (no more than a quarter of the size of the background image), and extract the inserted object manually or automatically.

(2) Then generate the edge image and energy map ^[2] of the background image, and select the best area from the background image based on these two information plus color difference information.

(3) Generate a line detection map of the background image, select the most relevant line as a reference line segment, and perform affine transformation processing on the inserted object.

(4) Finally, the method of Poisson fusion ^[1] is used to fuse the inserted object and the background image in the selected best area.

An experimental example of this method is shown in Figure 2. As shown in the figure, when the Ford advertisement icon is inserted, the lower edge of the window is selected as the most relevant straight line, and an affine transformation is performed on the icon according to the direction of the straight line. Therefore, the inserted icon is in harmony with the background wall, so that the icon can be inserted into the original picture naturally and eye-catchingly without affecting the quality of the original picture.

references:

[1] EREZ P. GANGNET M, BLAKE A. Poisson image editing [J]. ACM Transactions on Graphics, 2003, 22(3): 313—318.

[2]Harel J, Koch C, Perona P.Graph-based visual saliency[C].Advances inner information processing systems.2006:545-552.

[3]Burt P J, Adelson E H.A multiresolution spline with application to image mosaics[J].ACM Transactions on Graphics(TOG),1983,2(4):217-236.

[4]Jia J, Sun J, Tang C K, et al.Drag-and-drop pasting[C]//ACMTransactions on Graphics(TOG).ACM,2006,25(3):631-637.

Claims (5)

1. A kind of 1. method of advertising logo insertion digital picture, it is characterised in that comprise the following steps that：

   (1) best region of background image is chosen

   Advertising logo is embedded in digital picture, advertising logo is insertion object, and digital picture is background image；Choose background image In there is no that strong edge, energy are relatively low, the color regional area larger with insertion object color gap is as candidate region；It is and then logical Cross the colouring information for comparing insertion object and digital picture, the energy information and marginal information of background image, in background image Choose best region；Its best regionMeet equation below：

   <mfenced open = "" close = ""> <mtable> <mtr> <mtd> <munder> <mrow> <mi>m</mi> <mi>a</mi> <mi>x</mi> </mrow> <mi>i</mi> </munder> </mtd> <mtd> <mrow> <msub> <mi>V</mi> <mi>i</mi> </msub> <mo>=</mo> <msub> <mi>&amp;alpha;E</mi> <mi>i</mi> </msub> <mo>+</mo> <msub> <mi>&amp;beta;B</mi> <mi>i</mi> </msub> <mo>+</mo> <msub> <mi>&amp;gamma;C</mi> <mi>i</mi> </msub> </mrow> </mtd> </mtr> </mtable> </mfenced>

   Wherein：V_iRepresent the appropriate degree of ith zone in background image, be worth bigger expression region i as insert region effect more It is good；E_iRefer to the energy information of the background image of ith zone, energy information more meets selection standard, and the value is bigger；

   B_i,C_iMarginal information and color contrast information are represented respectively；α, β, γ are the weight parameters artificially chosen；

   (2) affine transformation of object is inserted

   Extracted first with the edge image of rim detection generation background image, then with the method for Hough transformation in background image Straight line, in the straight-line detection figure of background image, maximally related straight line is chosen, calculate the parameter of affine transformation, it is straight with this The direction of line is the affine transformation that standard insert object, i.e., the conversion in angle is carried out to insertion object so that insertion pair Best region as being more bonded background image；

   (3) image co-registration

   The insertion object obtained after affine transformation is melted with background image using based on the image interfusion method of Poisson's equation Close.
2. 2. according to the method for claim 1, it is characterised in that in the formula of step (1), α, beta, gamma is disposed as 0.33.
3. 3. according to the method for claim 1, it is characterised in that the regional area without strong edge is from edge image Find, that is, select edge strength to be less than the region of set threshold value, find more smooth region, prevent that inserting object crosses over background The multiple objects of image；The relatively low regional area of the energy is found from energy diagram, that is, selects energy value to be less than set threshold value Region, in order to avoid insertion object be placed to important objects, there is obvious artificial treatment vestige；The colour-difference away from compared with Big regional area directly contrasts to obtain by the colouring information of background image and insertion object, is in order that insertion object is tried one's best Eye-catching, its final selected region representation is RE.
4. 4. according to the method for claim 1, it is characterised in that maximally related when selecting the maximally related straight line Straight line is located at from the range of with a certain distance from selected areas RE, the certain distance be selected areas RE with nearest straight line away from From three times；The length of straight line and the distance from background best region are considered during selection, the straight line is from selected areas Nearer, correlation is bigger；Straight line is longer, more representative, and correlation is also bigger.
5. 5. according to the method for claim 1, it is characterised in that before the advertising logo insertion digital picture, to advertisement Icon is pre-processed；Pre-treatment step is as follows：

   (1) edge of advertising logo is handled so that do not include incoherent excess pixel in advertisement trademark；

   (2) advertisement trademark is zoomed in and out, after scaling, the size for inserting object is no more than a quarter of background image size.